A generalized linear models approach to spatial scan statistics for covariate adjustment

The spatial scan statistic proposed by Kulldorff (Commun. Statist.-Theory Methods 1997; 26:1481-1496) is one of the most widely used methods for detecting spatial clusters and evaluating their statistical significance. However, it is not fully capable of adjusting for all types of confounding covariates. In this article, a generalized linear models (GLM) approach to construct spatial scan statistics, which is readily in a form for covariate adjustment, is proposed. Using GLM, spatial scan statistics for different probability models call be formulated in a single framework. The test statistic is based on the log-likelihood ratio test statistic and evaluated using Monte Carlo hypothesis testing. The proposed method is illustrated using Texas female breast cancer data concerning late versus early stage cancer cases with covariates of race/ethnicity and age group. Copyright (C) 2009 John Wiley & Sons, Ltd.